fo.in rs de ea yr AI Course Lecture 1- 6, notes, slides www.myreaders.info/ , RC Chakraborty, e-mail
[email protected] , June 01, 2010 www.myreaders.info/html/artificial_intelligence.html ab or ty ,w w w .m Artificial Intelligence - Introduction : R C C ha kr www.myreaders.info Introduction Artificial Intelligence Introduction to Artificial Intelligence, topics : Definitions, goals, approaches, techniques, and branches; Intelligent behavior, understanding AI, hard or strong AI, soft or weak AI, cognitive science. General, engineering and science based AI Goals. AI approaches - cognitive science, laws of thought, turing test, rational agent. AI Techniques that make system to behave as intelligent describe and match, goal reduction, constraint satisfaction, tree searching, generate and test, rule based systems. Biology-inspired AI techniques - neural networks, genetic algorithms, reinforcement learning. Branches of AI - logical AI, search in AI, pattern recognition, knowledge representation, inferencing, common sense knowledge and reasoning, learning, planning, epistemology, ontology, heuristics, genetic programming. Applications of AI - game playing, speech recognition, understanding natural language, computer vision, expert systems. fo .in rs de ea ,w w w .m yr Introduction ha kr ab or ty Artificial Intelligence R C C Topics (Lectures 01, 02, 03, 04, 05, 06 6 hours) Slides 03-10 1. Definitions Artificial Intelligence, Intelligence, Intelligent behavior, Understanding AI, Hard or Strong AI, Soft or Weak AI, Cognitive Science. 2. Goals of AI 11-12 General AI Goal, Engineering based AI Goal, Science based AI Goal. 3. AI Approaches 13-16 Cognitive science, Laws of thought, Turing Test, Rational agent. 4. AI Techniques 17-32 Techniques that make system to behave as Intelligent Describe and match, Goal reduction, Constraint satisfaction, Tree Searching, Generate and test, Rule based systems, Biology-inspired AI techniques Neural Networks, Genetic Algorithms, Reinforcement learning. 5. Branches of AI 33-45 Logical AI, Search in AI, Pattern Recognition, Knowledge Representation, Inference, Common sense knowledge and reasoning, Learning, Planning, Epistemology, Ontology, Heuristics, Genetic programming. 6. Applications of AI 46-50 Game playing, Speech Recognition, Understanding Natural Language, Computer Vision, Expert Systems. 7. References 02 51 fo .in rs de ea yr What is Artificial Intelligence ? ha kr ab or ty ,w w w .m Introduction R C C • John McCarthy, who coined the term Artificial Intelligence in 1956, defines it as "the science and engineering of making intelligent machines", especially intelligent computer programs. • Artificial Intelligence (AI) is the intelligence of machines and the branch of computer science that aims to create it. • Intelligence is the computational part of the ability to achieve goals in the world. Varying kinds and degrees of intelligence occur in use of people, many animals and some machines. • AI is the study of the mental faculties through the computational models. • AI is the study of : How to make computers do things which, at the moment, people do better. • AI is the study and design of intelligent agents, where an intelligent agent is a system that perceives its environment and takes actions that maximize its chances of success. 03 fo .in rs de ea AI - Definitions 1.1 Artificial Intelligence (AI) ab or ty ,w w w .m yr 1. Definitions C C ha kr The definitions of AI outlined in textbooks exciting new effort to make (b) 'The study of mental faculties through computers think ... machines with the use of computational models' minds, in the full and literal sense' (Charniak and McDermott, 1985) (Haugeland, 1985) R (a) 'The 'The automation of activities that we associate with human thinking, activities such as decision-making, problem solving, learning ...' (Bellman, 1978) 'The study of the computations that make it possible to perceive, reason, and act' (Winston, 1992) (C) 'The art of creating machines that (d) 'A field of study that seeks to explain perform functions that require and emulate intelligent behavior in intelligence when performed by terms of computational processes' people' (Kurzweil, 1990) (Schalkoff, 1990) 'The study of how to make computers do things at which, at the moment, people are better' (Rich and Knight, 1991) ▪ 'The branch of computer science that is concerned with the automation of intelligent behavior' (Luger and Stubblefield, 1993) The definitions on the top, (a) and (b) are concerned with reasoning, whereas those on the bottom, (c) and (d) address behavior. ▪ The definitions on the left, (a) and (c) measure success in terms of human performance, whereas those on the right, (b) and (d) measure ideal concept of intelligence called rationality. Note : 04 A system is rational if it does the right thing. fo .in rs de ea AI - Definitions or ty ,w w w .m yr 1.2 Intelligence Relate to tasks involving higher mental processes. ha kr ab Examples: R C C creativity, solving problems, pattern recognition, classification, learning, induction, deduction, building analogies, optimization, language processing, knowledge and many more. Intelligence is the computational part of the ability to achieve goals. 05 R C C ha kr ab or ty . using analogies.fo . Reasoning to solve problems and discover hidden knowledge.Definitions Perceiving one’s environment. Communicating with others. Expressive-ness. Learning and understanding from experience.w w w . Acting in complex environments. Ingenuity.m yr 1. Knowledge applying successfully in new situations. Curiosity.in rs de ea AI . Thinking abstractly. and more like Creativity.3 Intelligent Behavior 06 . w w w . and plan for future. belief. dreams.” 07 . fear. emotions. hope. How mechanisms of intelligence produce the phenomena of illusion.m yr 1. represented.Definitions How knowledge is acquired. How motives. How sensory signals are transformed into symbols. to reason about past. and stored. kr ab or ty .4 Understanding AI R C C ha How intelligent behavior is generated and learned.in rs de ea AI . kindness and love. How symbols are manipulated to perform logic.fo . and priorities are developed and used. artificial intelligence research aims to create AI that can R C C ha kr replicate human intelligence completely. ◊ If it can apply a wide range of background knowledge and ◊ If it has some degree of self-consciousness. ▪ Strong AI refers to a machine that approaches or supersedes human intelligence.in rs de ea AI .5 Hard ▪ Generally.fo . ◊ If it can do typically human tasks.m yr 1. 08 . ▪ Strong AI aims to build machines whose overall intellectual ability is indistinguishable from that of a human being.Definitions or Strong AI ab or ty .w w w . it is merely an intelligent. 09 .m yr 1. ▪ Example : a chess program such as Deep Blue.fo .6 Soft or ▪ Weak AI refers to the use of software to study or accomplish specific C ha kr problem solving or reasoning tasks that do not encompass the full R C range of human cognitive abilities.Definitions Weak AI ab or ty . ▪ Weak AI does not achieve self-awareness.in rs de ea AI . a specific problem-solver.w w w . it demonstrates wide range of human-level cognitive abilities. but say little about the ways humans play chess. the program must operate in an intelligent manner.Definitions ▪ ab or ty . ▪ The important is not what is done but how it is done.7 Cognitive Science Aims to develop.w w w . explore and evaluate theories of how the mind R C C ha kr works through the use of computational models.fo .m yr 1. ▪ Example : The Chess programs are successful. means intelligent behavior is not enough. 10 .in rs de ea AI . Goals of AI • Traditionally. 3.w w w . Systems that act rationally R C C ha kr ab or ty .m yr 2. Systems that act like humans 4. Systems that think rationally.in rs de ea Goals of AI • The definitions of AI gives four possible goals to pursue : 1.fo . 2. all four goals have been followed and the approaches were: Rationally Human-like Think (1) Cognitive science Approach (2) Laws of thought Approach Act (3) Turing Test Approach (4) Rational agent Approach • Most of AI work falls into category (2) and (4). Systems that think like humans. 11 . Science based AI Goal Develop concepts. Engineering based AI Goal Develop concepts.in rs de ea Goals of AI or ty . Enhance human-human.fo . R C C Make an intelligent connection between perception and action. theory and practice of building intelligent machines Emphasis is on system building. mechanisms and vocabulary to understand biological intelligent behavior. ha kr ab Solve knowledge intensive tasks. 12 .m yr • General AI Goal Replicate human intelligence : still a distant goal. Emphasis is on understanding intelligent behavior. human-computer and computer to computer interaction / communication.w w w . the with minds. R C 3.1 Cognitive science : Think human-like ▪ An exciting new effort to make computers think. AI Approaches The approaches followed are defined by choosing goals of the computational C ha kr ab model. ▪ Computational model as to how results were obtained. ▪ Goal is not just to produce human-like behavior but to produce a sequence of steps of the reasoning process. but looks at reasoning process. ▪ Focus is not just on behavior and I/O.fo . in the full and literal sense.m yr 3.in rs de ea AI Approaches or ty .w w w . machines that it is. and basis for evaluating performance of the system. similar to the steps followed by a human in solving the same task. 13 . ” ▪ Goal is to formalize the reasoning process as a system of logical rules and procedures for inference.in rs de ea AI Approaches or ty .2 Laws of Thought : Think Rationally ▪ The study of mental faculties through the use of computational models. the study of the computations that make it possible to R C C ha perceive. reason. ▪ The issue is. ▪ Focus is on inference mechanisms that are provably correct and guarantee an optimal solution.m yr 3.fo . Therefore Socrates is mortal.w w w . kr ab that it is. and act. ▪ Develop systems of representation to allow inferences to be like “Socrates is a man. All men are mortal. 14 . not all problems can be solved just by reasoning and inferences. how to C C ha make computers do things which at the moment people do better. 15 . a computer. then conclude that the machine is intelligent. ▪ Goal is to develop systems that are human-like.3 Turing Test : Act Human-like kr ab intelligence when performed by people. and the person also tries to convince the interrogator that it is the human. ◊ The machine tries to fool the interrogator to believe that it is the human. ◊ The interrogator can communicate with the other 2 by teletype (to avoid the machine imitate the appearance or voice of the person). R ▪ Focus is on action.fo .m yr 3. ◊ If the machine succeeds in fooling the interrogator.in rs de ea AI Approaches ▪ The art of creating machines that perform functions requiring or ty . ▪ A Behaviorist approach. and not intelligent behavior centered around representation of the world. that it is the study of. ◊ The interrogator tries to determine which is the person and which is the machine.w w w . ▪ Example : Turing Test ◊ 3 rooms contain: a person. is not concerned with how to get results but to the similarity to what human results are. and an interrogator. m yr 3. that it is concerned with the automation of R C C ha intelligence. ▪ Goal is to develop systems that are rational and sufficient.fo .4 Rational Agent : Act Rationally Tries to explain and emulate intelligent behavior in terms of kr ab computational processes. 16 ▪ Focus is on systems that act sufficiently if not optimally in all situations.w w w .in rs de ea AI Approaches ▪ or ty . . ▪ It is passable to have imperfect reasoning if the job gets done. kr ab The techniques are concerned with how we represent. manipulate and reason R C C ha with knowledge in order to solve problems. not all "intelligent" but used to behave as intelligent Describe and match Goal reduction Constraint satisfaction Tree Searching Generate and test Rule based systems • Biology-inspired AI techniques are currently popular 17 Neural Networks Reinforcement learning Genetic Algorithms .w w w . AI Techniques Various techniques that have evolved.fo .m yr 4. can be applied to a variety of AI tasks. Example • Techniques.in rs de ea AI Techniques or ty . it is called non-deterministic if at least one state has more than one successor. a set of start states.w w w . ■ Computation model is a finite state machine.m yr 4. S') is such that one move takes the system from S to S'. R ■ Finite state model consists of a set of states. ■ Examples of some possible transitions between states are shown for the Towers of Hanoi puzzle. ■ Representation of computational system include start and end state descriptions and a set of possible transition rules that might be applied. It includes of a set of states.1 Techniques that make system to behave as "Intelligent" C C ha kr ■ Model is a description of a system’s behavior. then the transition relation is represented by S => S’ ■ State-transition system is called deterministic if every state has at most one successor. a set of input events and the relations between them. an input alphabet.in rs de ea AI Techniques • Describe and Match ab or ty . Given a current state and an input event you can determine the next current state of the model. and a transition function which maps input symbols and current states to a next state.fo . 18 . Problem is to find the appropriate transition rules. ■ Transition relation: If a pair of states (S. Play the game in the smallest number of moves possible . 2] [ ] Towers of Hanoi puzzle ■ Shortest solution is the sequence of transitions from the top state downward to the lower left.in rs de ea yr ■ Puzzle : Towers of Hanoi with only 2 disks Solve the puzzle : ab or ty .w w w . middle post can be used for intermediate storage.fo .m AI Techniques Goal state R C C ha kr Initial state Move the disks from the leftmost post to the rightmost post while never putting a larger disk on top of a smaller one. from one peg to another. 19 . 2] [] [1 ] [1] [2] [ ] [2] [] [2] [1] [] [1] [1] [2] [ ] [ ] [1. [1. move one disk at a time. ■ Possible state transitions in the Towers of Hanoi puzzle with 2 disks. 2] [ ] [ 2] [1 ] [] [2] [] [] [ ] [1. and lowerlevel goals are lower in the tree. ■ The process involves the hierarchical sub-division of goals into subgoals.w w w . ◊ Goal levels : Higher-level goals are higher in the tree. ◊ Nodes at the bottom of the tree represent irreducible action goals.fo . alternative sub-goals and conjoint sub-goals. ◊ Arcs are directed from a higher-to-lower level node represents the reduction of higher-level goal to lower-level sub-goal.in rs de ea AI Techniques ab or ty . ■ Goal-reduction process is illustrated in the form of AND/OR tree drawn upside-down. ■ An AND-OR tree/graph structure can represent relations between goals and sub-goals. until the sub-goals which have an immediate solution are reached and said “goal has been satisfied”.m yr • Goal Reduction ■ Goal-reduction procedures are a special case of the procedural ha kr representations of knowledge in AI. logic- R C C based representations. an alternative to declarative. [example in the next slide] 20 . C ha kr ab “earning/save money”.g. is a sub-goal of “improving enjoyment of life”.in rs de ea yr ■ Example Goal Reduction or ty . but as well need to “save money”. not only need to “earn more money”. is a sub-goal of “improving standard of living ”. ◊ Conjoint sub-goals To “provide for old age”.w w w . e. ◊ Alternative ways of trying to solve a goal The “going on strike” and “increasing productivity” are alternative ways of trying to “earn more money” (increase pay). R C Improve enjoyment of life OR Improve standard of living Work less hard Provide for old age AND Save money Earn more money OR Improve productivity Go on strike AND-OR tree/graph structure The above AND-OR tree/graph structure describes ◊ Hierarchical relationships between goals and subgoals The “going on strike” is a sub-goal of “earning more money”. 21 .fo .: “improving standard of living” and “working less hard” are alternative ways of trying to “improve enjoyment of life”.m AI Techniques AND-OR tree/graph structure to represent facts such as “enjoyment”. “old age” etc. . ■ Constraint Satisfaction Problem (CSP) and its solution ◊ A Constraint Satisfaction Problem (CSP) consists of : ‡ Variables. ‡ Constraints. the relation is still maintained.w w w . . or a good solution . 22 . ‡ all solutions.g. moving any one.fo . a set of values that the variables can simultaneously satisfy the constraint (e. that could be : ‡ one solution. D1 != D2) ◊ A solution to a CSP is an assignment of a value from its domain to every variable satisfying every constraint.in rs de ea AI Techniques ab or ty .m yr • Constraint Satisfaction Techniques ■ Constraint is a logical relation among variables.Constraint Optimization Problem (COP). a finite set Di of possible values which each variable xi can take. “circle is inside ha kr the square” – The constraints relate objects without precisely specifying R C C their positions. a finite set X = {x1 . e. ‡ Domain. ‡ an optimal.g. . with no preference as to which one. ■ Constraint satisfaction is a process of finding a solution to a set of constraints – the constraints express allowed values for variables and finding solution is evaluation of these variables that satisfies all constraints. . xn } . ◊ Apply "no-threatening" constraints between each couple Ri and Rj of queens and evolve the algorithm. (a queen threatens other queens on same row. 23 . column and diagonal). ◊ Assign a variable Ri (i = 1 to N) to the queen in the i-th column indicating the position of the queen in the row.in rs de ea yr . If solutions that differ only by symmetry operations (rotations and reflections) of the board are counted as one.w ty ■ Constraint satisfaction has application in Artificial Intelligence. ◊ Example : 8 .Queens puzzle a b c d e f g h a b c d e f g h 8 8 8 8 7 7 7 7 6 6 6 6 5 5 5 5 4 4 4 4 3 3 3 3 2 2 2 2 1 1 1 1 a b c d e f g h a Unique solution 1 b c d e f g h Unique solution 2 ◊ The eight queens puzzle has 92 distinct solutions. Solution : To model this problem ◊ Assume that each queen is in different column.m w w . Programming Languages. Only two solutions are presented above. place N queens on N*N chessboard satisfying constraint that no two queens threaten each other.fo . kr ab or AI Techniques [continued from Constraint Satisfaction Techniques] R C C ha ■ Example 1 : N-Queens puzzle Problem : Given any integer N. Computational Logic. The puzzle has 12 unique solutions. Symbolic Computing. 24 . ◊ Introduce the non-equality constraint between two variables labeling adjacent nodes. ◊ A 4 – color map ◊ The "Four Color Theorem". states that 4 . the problem is kr ab to assign colors to those areas in the map (nodes) satisfying the C C ha constraint that no adjacent nodes (areas) have the same color assigned R to them.w w w .colors are sufficient to color any map so that regions sharing a common border receive different colors.m AI Techniques Problem : Given a map (graph) and a number of colors. Solution : To model this Map Coloring problem ◊ Label each node of the graph with a variable (domain corresponding to the set of colors).fo .in rs de ea yr ■ Example 2 : Map Coloring or ty . and test this path. "most likely" path to follow at each node.backtrack to the last node in the tree where a previously unexplored path branches of.m yr • Tree Searching ■ Many problems (e.w w w . * In the case of a dead end . R ■ A search through the entire tree. The four other types of search strategies are : ◊ Hill climbing ◊ Breadth-first search * Like depth first but involving * Look for a solution amongst some quantitative decision on all nodes at a given level the before proceeding to the next. "most likely" to lead to a solution.in rs de ea AI Techniques or ty . * At each node. * Backtracking can be of two types : – Chronological backtracking : undo everything as we move back "up" the tree to a suitable node. – Dependency directed backtracking : only withdraw choices that "matter" (ie those on which dead end depends).g. goal reduction. ◊ Beam search * Like breadth first (level by * Like beam search but only level) but selecting only those proceeding from one "most N nodes at each level that are likely" node at each level. constraint networks) can be kr ab described in the form of a search tree. pick an arbitrary path and work forward until a solution is found or a dead end is reached. ■ Tree search strategies: ◊ Depth-first search * Assumes any one path is as good as any other path. A solution to the problem is C C ha obtained by finding a path through this tree.fo . is called exhaustive search. until a satisfactory path is found. 25 ◊ Best-first search . 26 . ◊ For better efficiency GT approach need to be supported by backtracking approach. generates many wrong assignments of values to variables which are rejected in the testing phase. ◊ Generator leaves conflicting instantiations and it generates other assignments independently of the conflict. ◊ This paradigm involves two processes: * Generator to enumerate possible solutions (hypotheses).in rs de ea AI Techniques or ty .fo . * Test to evaluate each proposed solution ◊ The algorithm is Generate labels Test satisfaction Disadvantages ◊ Not very efficient. or to R prove that the problem is unsolvable. Example: Opening a combination lock without knowing the combination. C C ha ◊ CSP algorithms guarantee to find a solution. ■ Generate-and-test method The method first guesses the solution and then tests whether this solution is correct. if one exists.w w w . means solution satisfies the constraints.m yr • Generate and Test (GT) ■ Most algorithms for solving Constrain Satisfaction Problems (CSPs) kr ab search systematically through the possible assignments of values. ◊ the disadvantage is that they take a very long time to do so. Rule Base.m yr • Rule-Based Systems (RBSs) IF <condition> THEN <action>. ab or ty . Interpreter.in rs de ea AI Techniques ■ Rule-based systems are simple and successful AI technique. C ha kr ◊ Rules are of the form: R C ◊ Rules are often arranged in hierarchies (“and/or” trees). ■ RBS Components : Working Memory. ◊ When all conditions of a rule are satisfied the rule is triggered.fo . Change Interpreter Conditions Observed Data Rules Working Memory Rule Base Rule-Based System 27 Each of these 3 components are described in the next slide .w w w . Description or ty . 20> THEN <add (ocean. yes)> ‡ If the conditions are matched to the working memory and if fulfilled then rule may be fired. ‡ It selects rule from Rule Base and applies by performing action.in rs de ea yr ■ RBS components . over. ‡ Rule syntax is IF <condition> THEN <action> e. “Remove” fact(s) from WM. < car. color. ‡ Rules are domain knowledge and modified only from outside.m AI Techniques ◊ Working Memory (WM) kr ab ‡ Contains facts about the world observed or derived from a rule. ◊ Rule Base (RB) ‡ RB contains rules. 28 . reorders and resolves conflicts. attribute.w w w . ‡ Can be modified by the rules. ‡ Contains temporary knowledge about problem-solving session. ‡ RB actions are : “Add” fact(s) to WM. values > R e.Prunes. then applies the rule by performing action. red > : “The color of my car is red”. ◊ Interpreter ‡ It is the domain independent reasoning mechanism for RBS.Finds the rules that matches the current WM. swimable. Refinement .Executes the actions of the rules in the Conflict Set.fo . Execution . each rule is a step in a problem solving.g. ‡ It operates on a cycle: Retrieval . IF <(temperature.g. “Modify” fact(s) in WM. C C ha stored as a triplet < object. .. T wn Xn Y ◊ w1. . w2.m yr 4. wn are weights of the edges and are real valued.fo . they train weights of "neurons". in which inputs and outputs are known and the goal is to build a representation of a function that will approximate the input to output mapping.2 Biology-Inspired AI Techniques C ha kr ■ Neural Networks model a brain learning by example. R C ■ Neural networks are structures trained to recognize input patterns. xn are inputs as real X1 w1 w2 X2 X3 numbers or boolean values depends w3 ● ● on problem.. If the net input which is w1 x1 + w2 x2 + .w w w .in rs de ea AI Techniques • Neural Networks (NN) ab or ty .. 29 . Perceptron Model ◊ y is the output and is boolean. inside... ■ A Perceptron is a model of a single `trainable' neuron. shown below : ◊ x1.. ◊ T is the threshold and is real valued. ■ Neural networks typically take a vector of input values and produce a vector of output values. . x2.. + wn xn is greater than the threshold T then output y is 1 else 0. ■ Neural networks use supervised learning. in rs de ea AI Techniques ■ GAs are part of evolutionary computing. ◊ A gene may be mutated and expressed in the organism as a completely new trait. setting for a hair color gene may be black or brown. and encoded in the genes of an organism. the nature uses. 30 . a rapidly growing area of AI. ◊ The genes and their settings are referred as an organism's genotype. ◊ When two organisms mate they share their genes. ◊ Each gene represents a specific trait (feature) of the organism and has several different settings. ■ Thus.m yr • Genetic Algorithms (GA) ha kr ■ Genetic algorithms are implemented as a computer simulation.w w w . This process is called cross over. Crosses over. where R C C techniques are inspired by evolutionary biology. The resultant offspring may end up having half the genes from one parent and half from the other. ■ Mechanics of biological evolution ◊ Every organism has a set of rules. Genetic Algorithms are a way of solving problems by mimicking processes. describing how that organism is built. Selection. e.g.fo . ab or ty . Mutation and Accepting to evolve a solution to a problem. ◊ The genes are connected together into long strings called chromosomes. cross over the parents to form new offspring (children).m AI Techniques (1) [Start] Generate random population of n chromosomes (Encode kr ab suitable solutions for the problem) C C ha (2) [Fitness] Evaluate the fitness f(x) of each chromosome x in the R population. If no crossover was performed. stop. (5) [Test] If the end condition is satisfied. (d) [Accepting] Place new offspring in the new population. (b) [Crossover] With a crossover probability. mutate new offspring at each locus (position in chromosome). 31 . (3) [New population] Create a new population by repeating following steps until the new population is complete. and return the best solution in current population. (6) [Loop] Go to step 2. (a) [Selection] Select two parent chromosomes from a population according to their fitness.in rs de ea yr ■ Genetic Algorithm Steps or ty . offspring is the exact copy of parents. (4) [Replace] Use new generated population for a further run of the algorithm.w w w . ■ Genetic Algorithms does unsupervised learning – the right answer is not known beforehand.fo . (c) [Mutation] With a mutation probability. the agent develops a policy π : S A which maximizes the quantity R = r0+ r1+.in rs de ea AI Techniques or ty . ◊ Based on these interactions.w w w . ■ The decision-making agent interacts with its environment so as to maximize the cumulative reward it receives over time. ◊ It chooses an action a ∈ A(st) and receives from the environment the new state st+1 and a reward rt+1.S ” . 32 .A”. The steps are: ◊ At each time t. C C ha ■ RL is conducted within the mathematical framework of Markov decision R processes (MDPs). and a set of scalar “rewards . everyday lives. ■ RL methods focus on the kind of learning and decision making problems that people face in their normal. the agent perceives its environment state st and the set of possible actions A(st) .R”. +rn for MDPs. from the kr ab consequences of action. ■ The basic RL model consists : a set of “environment states .m yr • Reinforcement Learning (RL) ■ RL is learning from interaction with an environment.fo . rather than from explicit teaching. a set of “actions . . Truth value “ true”. either true or false but not both. ■ Types of logic ◊ Propositional logic .dealing with fuzziness ◊ Temporal logic. etc ■ Propositional logic and Predicate logic are fundamental to all logic ◊ Propositional logic ‡ Propositions are “Sentences”.fo . else “false” ‡ Example : Sentence "Grass is green". Branches of AI ha kr ■ Logic is a language for reasoning. Object represented as x .in rs de ea Branches of AI • Logical AI ab or ty . Proposition is “yes” ◊ Predicate logic ‡ Predicate is a function may be true or false for arguments ‡ Predicate logic are rules that govern quantifiers ‡ Predicate logic is propositional logic added with quantifiers ‡ Examples: “The car Tom is driving is blue". "The sky is blue".m yr 5. ‡ A sentence is smallest unit in propositional logic ‡ If proposition is true. read B(x) as "x is blue" . "The cover of this book is blue" 33 Predicate is blue. Sentence represented as B(x).logic of objects ◊ logic involving uncertainties ◊ Fuzzy logic .logic of sentences ◊ predicate logic . give a name B . a collection of rules used while R C C doing logical reasoning. then truth value is "true".w w w . . Conditions for goal satisfaction ◊ Path cost .Successor function : reachable states ◊ Goal test .optimal if lowest cost.fo . the most fundamental are ◊ Depth first ◊ Hill climbing ◊ Breadth first ◊ Least cost ■ Search components ◊ Initial state .find a sequence of actions.m yr • Search in AI ■ Search is a problem-solving technique that systematically consider R C C ha kr ab all possible action to find a path from initial state to target state.Cost of sequence from initial state to reachable state ■ Search objective ◊ Transform initial state into goal state . ■ Search solution ◊ Path from initial state to goal .First location ◊ Available actions . ■ Search techniques are many.w w w . 34 .in rs de ea Branches of AI or ty . fo . ATR ◊ Character recognition – Mail sorting.w w w . access ■ Approaches for Pattern recognition ◊ Template Matching ◊ Statistical classification ◊ Syntactic or Structural matching Each of 35 these approaches are explained in the next slide.Medical image/EEG/ECG signal analysis ◊ Speech recognition .in rs de ea Branches of AI ■ Definitions : from the literature ab or ty . processing bank cheques ◊ Computer aided diagnosis .Human Computer Interaction. .m yr • Pattern Recognition (PR) ha kr ◊ 'The assignment of a physical object or event to one of pre-specified R C C categories' – Duda and Hart ◊ 'The science that concerns the description or classification (recognition) of measurements' – Schalkoff ◊ 'The process of giving names Ω to observations X ' – Schürmann ◊ Pattern Recognition is concerned with answering the question 'What is this?' – Morse ◊ 'A problem of estimating density functions in a high-dimensional space and dividing the space into the regions of categories or classes' – Fukunaga ■ Pattern recognition problems ◊ Machine vision .Visual inspection. primitives are viewed as the alphabet of the language. and adapt themselves to input data. measure similarity (correlation) based on training set.in rs de ea yr .m w w . The patterns are viewed as sentences belonging to a language. they themselves are built from simpler / elementary sub-patterns are called primitives. A large collection of complex patterns can be described by a small number of primitives and grammatical rules.w ty Branches of AI [continued from previous slide Approaches for Pattern Recognition ] Template Matching kr ab or Match with stored template considering translation.fo . Statistical classification Each pattern is represented in terms of d features (measurements) and viewed as a point in a d-dimensional space. The sentences are generated according to a grammar.following decision theoretic or discriminant analysis approaches. rotation and scale R C C ha changes. Syntactic or Structural matching Complex pattern is composed of sub-patterns and the relations. Neural networks Neural networks are viewed as weighted directed graphs in which the nodes are artificial neurons and directed edges (with weights) are connections between neurons input-output. ■ Applications requiring Pattern recognition ◊ Image Proc / Segmentation ◊ Seismic Analysis ◊ Computer Vision ◊ Industrial Inspection ◊ Medical Diagnosis ◊ Financial Forecast ◊ Man and Machine Diagnostics 36 . The grammar for each pattern class are inferred from the training samples. Neural networks have the ability to learn complex nonlinear input-output relationships from the sequential training procedures. Using training sets establish decision boundaries in the feature space . The Scripts are linked sentences using frame-like structures. default. and the arcs represent binary relationships between concepts.w w w . To manipulate these facts by a C C ha program. Instantiation Procedure (inheritance. ◊ Production rules : Consists of a set of rules about behavior. ◊ Semantic networks : A semantic net is just a graph. e.. 37 . a suitable representation is required. then the production is said to be triggered. ◊ Frames and scripts : A frame is a data structure that typically consists of : Frame name. ■ Knowledge representation formalisms (techniques) Different types of knowledge require different types of representation.g. where the nodes represent concepts. A Good representation R facilitates problem solving. Pointers (links) to other Frames.fo . a record of sequence of events for a given type of occurrence.m yr • Knowledge Representation ■ How do we represent what we know? kr ab Knowledge is a collection of facts. and FALSE for others. consistency). if a production's precondition matches the current state of the world. a production consists two parts: a precondition (or IF) and an action (or THEN). Slot-filler (relations target).in rs de ea Branches of AI or ty . ◊ Predicate logic : Predicate is a function may be TRUE for some arguments. it is deduction of new facts from old ones. ■ Inductive Inference ◊ It may be correct or incorrect inference. but this conclusion can be reversed when we hear that it is a penguin. 38 . ◊ A traditional logic is based on deduction. if the truth of a proposition does not change when new information are added to the system. there is no possibility of mistake if rules are followed exactly. also called Induction or Inductive reasoning. we human infer that bird can fly. Logic R C C ha captures inference. ■ Deductive Inference ◊ It is never false. it is a method of exact inference.m yr • Inference Inference is the act or process of deriving a conclusion based solely on what kr ab one already knows. and consistent . inexact.fo . the bird penguin cannot fly. if the process of reasoning in which the premises of an argument are believed to support the conclusion but do not ensure it.in rs de ea Branches of AI or ty . because in real world the information is incomplete.w w w . ◊ The reasoner draws conclusions tentatively reserving the right to retract them in the light of further information. ◊ A logic is non-monotonic. ◊ A logic is inductive. ◊ A logic is monotonic. inconsistent. if the truth of a proposition may change when a new information (axioms) is added to or an old information is deleted from the system. ◊ Example: When we hear of a bird. precise. ◊ The information required is complete. inference is true if premise is true. ◊ Project “Cyc” – It is an attempt to manually build a database of human common sense knowledge. ■ Teaching computers common sense ◊ Project “OpenMind” at MIT . ab or ty . and using such knowledge. ◊ Computer can not think.m yr • Common Sense Knowledge and Reasoning C ha kr ■ It is the ability to analyze a situation based on its context.fo . the computers programs do many things. R C ◊ People can think because the brain contain vast libraries of common sense knowledge and has means for organizing. it has 1. acquiring. ■ Researchers have divided common sense capability into : ◊ Common sense knowledge and ◊ Common sense reasoning. but still far away from several hundred million needed. and Story scripts. Currently. 39 .Here the goal is to teach a computer things that human take them for granted. they can play chess at the level of best players but cannot match capabilities of a 3 year old child at recognizing objects. here the knowledge is represented in the form of Semantic net.5 million collection of common sense facts. Probabilistic graphical models. computers lack common sense.w w w .in rs de ea Branches of AI Common sense is the mental skills that most people have. fo . ◊ Analogy : Learning from similarities. mutation and accepting to evolve a solution to a problem." ◊ Mitchell 1997 – “A computer program is said to learn from experience E with respect to some class of tasks T and performance measure P. +ve or - ve. selection.w w w . improves with experience E. it learns which actions are good or bad. crosses over.m yr • Learning ha kr ■ Definitions R C C ◊ Herbert Simon 1983 – “Learning denotes changes in the system that are adaptive in the sense that they enable the system to do the same task or tasks more efficiently and more effectively the next time. Assign rewards. nature uses. if its performance at tasks in T. ◊ Reinforcement : Learning from actions. can determine correspondence between two different representations.” ◊ Ryszard Michalski 1986 – "Learning is constructing or modifying representations of what is being experienced. ◊ Genetic Algorithms : Learning by mimicking processes nature uses.” ■ Major Paradigms of Machine Learning ◊ Rote : Learning by memorization. ◊ Induction : Learning by example. Part of evolutionary computing. Process of learning by example where a system tries to induce a general rule from a set of observed instances. a way of solving problems by mimicking processes.” ◊ Marvin Minsky 1986 – “Learning is making useful changes in the working of our mind. Recognize similarities in information already stored. at the end of a sequence of steps. Saving knowledge so that it can be used again. as measured by P. 40 .in rs de ea Branches of AI Programs learn from what the facts or the behaviors can represent. ab or ty . fo . particularly ◊ facts about the effects of actions. kr ab Planning is a problem solving technique. ◊ resolving goal conflicts.w w w . From facts the program generate a strategy for achieving the goal.m yr • Planning A plan is a representation of a course of action. ■ Benefits of planning ◊ reducing search. and ◊ statement of a goal. ■ Strategy for planning A strategy is just a sequence of actions. and ◊ providing a basis for error recovery. R C C ha Planning is a reasonable series of actions to accomplish a goal. ■ Planning programs Start with facts about the world.in rs de ea Branches of AI or ty . 41 . ◊ facts about the particular situation. ab or ty . how to ride a bicycle). Propositions Truths Knowledge Beliefs ◊ Consider a knowledge of proposition ..w w w . ◊ knowing someone in person.g.fo . and S is justified in believing that P is true. and ◊ knowing a place or a city.a schema ‘S knows that P’ . Gettier claimed that the above account of knowledge is insufficient.m yr • Epistemology C ha kr ■ There are various kinds of knowledge: R C ◊ knowing how to do something (e. ■ Epistemology is the study of the kinds of knowledge that are required for solving problems in the world.in rs de ea Branches of AI Epistemology is the theory of knowledge. ◊ Question asked .what are the necessary and sufficient conditions? S knows that P if and only if P is true. 42 . ■ Epistemology is the study of knowledge and justified belief. S believes that P is true. or parameters that objects can have and share. collections. 43 . ◊ Relations: ways the objects can be related to one another. ◊ Attributes: properties. represents a domain and is used to reason about the objects in that domain and the relations between them.fo . characteristics. ■ Ontology is a specification of a conceptualization. features. or types of objects.w w w . ■ Ontology generally describe: ◊ Individuals (instances): the basic or ground level objects ◊ Classes: sets. as a form of knowledge representation about the world or some part of it.m yr • Ontology Ontology is concerned with existence. ■ Ontology is used in artificial intelligence. R C ■ Ontology is a data model.in rs de ea Branches of AI or ty . a study of the categories of things C ha kr ab that exist or may exist in some domain. the heuristic functions are : ◊ used to measure how far a node is from goal state. R C In computer science. a heuristic is an algorithm with provably good run times and with provably good or optimal solution. efficient rules. potentially at the cost of accuracy or precision.in rs de ea Branches of AI Heuristics are simple. come to judgments. ■ People use heuristics to make decisions. ◊ used to compare two nodes.fo . when facing complex problems or incomplete information. ab or ty . and solve problems. 44 . find if one is better than the other. ■ In AI programs.m yr • Heuristics C ha kr ■ Heuristics are in common use as Rule of thumb. ■ Heuristics are intended to gain computational performance or conceptual simplicity. These rules work well under most circumstances.w w w . in rs de ea Branches of AI or ty . ◊ Controlling parameters for the run. and constants) ◊ Set of primitive functions ◊ Fitness measure (for fitness of individuals in the population).fo .m yr • Genetic programming (GP) Genetic programming is an automated method for creating program C ha kr ab from a high-level problem statement. functions.w w w . and ◊ Termination criterion for designating the result of the run. ■ The user (human) communicates the high-level statement of the problem to the GP system by performing certain well-defined preparatory steps. ■ The “Run” of genetic programming executes a series of well-defined problem-independent steps (the flowchart). the human user require to specify for the GP to be evolved are : ◊ Set of terminals (variables . ■ The major five preparatory steps. R C ■ GP starts from a high-level statement of the requirements of a problem and attempts to produce a computer program that solves the problem. 45 . w w w .m yr 6. partners. e. Chess. an emerging area in which R C C the goals of human-level AI are pursued. must approximate. opponents are unlikely to find goal. examples : Checkers. example : Go. ■ Computer Games ◊ Computers perform at champion level games. ■ Game play is a search problem defined by: ◊ Initial state .in rs de ea Applications of AI • Game playing ab or ty .fo .board ◊ Expand function . Applications of AI ha kr ■ Games are Interactive computer program.payoff of the state ◊ Goal test . and support characters that act just like humans.games become boring if there is no action for too long a time. Backgammon. ◊ Time limits . enemies. ◊ Computers perform well games.build all successor states ◊ Cost function . example : Bridge ◊ Computers still do badly.g. 46 Chess program won over world champion Gary . Othello. ■ Games are made by creating human level artificially intelligent entities. Hex ■ The Deep Blue Kasparov.ultimate state with maximal payoff ■ Game playing is characterized by: ◊ "Unpredictable" opponent ◊ Need to specify move for every possible opponent reply. w w w . computer speech recognition reached a practical level for R C C limited purposes.fo . ◊ Call routing (collect call). 47 . ◊ Speaker recognition. ■ The spoken language interface PEGASUS in the American Airlines' EAASY SABRE reservation system. ab or ty . allows users to obtain flight information and make reservations over the telephone. ■ The typical usages are : ◊ Voice dialing (Call home). ■ Using computers recognizing speech is quite convenient. ◊ Data entry (credit card number). but most users find the keyboard and the mouse still more convenient.m yr • Speech Recognition ha kr ■ In 1990s.in rs de ea Applications of AI ■ A process of converting a speech signal to a sequence of words. ■ Some major tasks in NLP ◊ Text-to-Speech (TTS) system : converts normal language text into speech.fo .in rs de ea Applications of AI or ty . ◊ Speech recognition (SR) system : process of converting a speech signal to a sequence of words. ◊ Machine translation (MT) system : translate text or speech from one natural language to another. ◊ Information retrieval (IR) system : search for information from databases such as Internet or World Wide Web or Intranets.w w w .m yr • Understanding Natural Language Natural language processing (NLP) does automated generation and C ha kr ab understanding of natural human languages. R C ■ Natural language generation system Converts information from computer databases into normal-sounding human language ■ Natural language understanding system Converts samples of human language into more formal representations that are easier for computer programs to manipulate. 48 . m yr • Computer Vision ■ It is a combination of concepts.C. ■ Examples ◊ Face recognition : the programs in use by banks ◊ Autonomous driving : The ALVINN system. but full computer vision requires partial 3-D information that is not just a set of 2-D views. and they are not as good as what humans evidently use. and in all weather conditions. Pattern Recognition. ◊ Other usages Handwriting recognition. autonomously drove a van from Washington. D. ■ At present there are only limited ways of representing 3-D information directly. ■ Some useful programs can work solely in 2-D. 49 . Artificial Intelligence and R C C ha Computer Graphics. ■ The world is composed of 3-D objects.fo . Manufacturing inspection. averaging 63 mph day and night. techniques and ideas from : Digital kr ab Image Processing.w w w . Baggage inspection. Photo interpretation. but the inputs to the human eye and computers' TV cameras are 2-D.in rs de ea Applications of AI or ty . to San Diego. etc . ■ Knowledge base A knowledge engineer interviews experts in a certain domain and tries to embody their knowledge in a computer program for carrying out some task. Insurance companies to assess the risk presented by the customer.g.g.w w w . ■ Expert systems rely on knowledge of human experts. etc.in rs de ea Applications of AI Systems in which human expertise is held in the form of rules ab or ty .fo . The "expertise" consists of knowledge about a particular domain. which diagnosed bacterial infections of the blood and suggested treatments. e. airline scheduling of flights ◊ Financial Decision Making : advisory programs assists bankers to make loans. ◊ Diagnosis and Troubleshooting : deduces faults and suggest corrective actions for a malfunctioning device or process ◊ Planning and Scheduling : analyzing a set of goals to determine and ordering a set of actions taking into account the constraints.m yr • Expert Systems C ha kr ■ It enable the system to diagnose situations without the human expert R C being present. and controlling optimality and do failure correction. ■ A Man-machine system with specialized problem-solving expertise. e. 50 . predicting trends. understanding of problems within that domain. noticing anomalies. and "skill" at solving some of these problems. ◊ Process Monitoring and Control : analyzes real-time data. ■ One of the first expert systems was MYCIN in 1974. Page 1-650. page 1-743. page 1-608. "Artificial Intelligence: A Modern Approach" by Stuart Russell and Peter Norvig. Morgan Kaufmann Inc. C ha 2. Page 1-493. Oxford University Press. 5. References : Textbooks 1. 51 An exhaustive list is . Chapter 1-22. 3. page 1-613. Related documents from open source.in rs de ea AI. (1994). by Thomas Dean..w w w . (1998). by George F.fo . Nilsson. by David Poole. "Artificial Intelligence".m yr 7. Chapter 1-27. Alan Mackworth. 7. Addison-Wesley. mainly internet. (2006). 4. (2002). Luger. Chapter 1-12. (1998). by Elaine Rich and Kevin Knight. "Artificial Intelligence: Structures and Strategies for Complex Problem Solving".16.Introduction . by Nils J. "Computational Intelligence: A Logical Approach". Chapter 1-10. page 1-1057. 6. R C (2002). and Randy Goebel. Prentice Hall. "Artificial Intelligence: Theory and Practice".. Chapter 1-25. Addison-Wesley.References ty . Chapter 1. being prepared for inclusion at a later date. McGraw Hill kr ab or companies Inc. "AI: A New Synthesis".